Sandor Vajda - US grants

The major component of the proposal is the development of methods designed to substantially increase the range of biomolecular systems to which advanced computational docking and design methodologies can be applied. The most successful methods and applications to date have been limited to proteins whose crystal structures are known, and which remain relatively rigid during complex formation. The main obstacle limiting the analysis of more general systems in which side chains and backbones change conformation is the target function. The proposal therefore focuses on methods for obtaining accurate, rapidly evaluatable semi-empirical free energy functions, with particular emphasis on solvation. These include methods for eliminating the time intensive surface area calculations required by current semi-empirical free energy functions, as well as several new approaches to solvation. Docking algorithms that can take full advantage of free energy- target functions will also be developed and tested in a variety of applications, exploiting the extensive crystallographic and thermodynamic databases available. Attention will be given to the systematic comparison of currently available and emerging semi-empirical free energy evaluation procedures, including the delineation of their range of applicability. J

This project focuses on two fundamental problems. The first is the development of flexible protein docking algorithms for the case in which the association occurs with substantial conformational change involving side chains and loop regions. The second problem is the development of con-formational search algorithms for refining low resolution protein models with 6 to 10 A RMSD from the native structure, obtained by homology modeling, fold recognition. or ab initio structure prediction. In both applications, free energy potentials, combining molecular mechanics with em-pirical solvation and entropic terms, will be used as target functions. It has been shown that such potentials can be more effective in decoy discrimination than purely empirical functions. However even with a normalization procedure to reduce the noise in the van der Waals energy, the combined potentials exhibit a multicomponent/multifrequency behavior, and are very difficult to minimize. The free energy is regarded as the sum of three functions: a smooth component that includes elec-trostatic, solvation, and entropic contributions, an intermediate frequency component of internal (bonded) energy terms, and the van der Waals term which is essentially a high frequency noise, and carries little information about the distance from the native state.

Most established methods of finding the global minimum of a multicomponent/multifrequency potential generate a large number of trial conformations and then rank them according to their free energies. This project will use a radically different approach that mimics native folding or association pathways. The basic idea of the method is performing minimization of the smooth free energy components, but restricting the search to regions of the conformational space with low van der Waals energy. Since this group has already shown that the combined potential has good discriminatory power, the key to a successful docking or model refinement is the ability to generate enough near-native conformations. It is expected that exploiting the gradient of the smooth free energy components will substantially increase the efficiency of sampling, and the resulting methods will be useful to the broad molecular biology/biochemistry community.

0002127
Vajda

Description: This award is to support the collaboration of Dr. Sandor Vajda, Department of Biomedical Engineering, Boston University, Boston, Massachusetts and Dr. Ugur Sezerman, Department of Biological Sciences and Bioengineering, Sabanci University, Tuzla, Istanbul, Turkey. Their goal is to further develop and apply docking and binding free energy computations to linear molecules interacting with protein targets, with specific emphasis on complex formation between antigenic peptides and major histocompatibility complex (MHC) proteins. They plan a three-step effort: Initially, to develop free energy potentials, by essentially combining molecular mechanics with empirical solvation and entropic terms. These would be used as target functions. The potentials they would obtain are to be tested on a set of protein-peptide docked decoys. In the second step, the (validated) new potentials will be used as a target function for peptide docking. This will be performed both by site mapping and via fragment assembly docking methods. The PIs intend to achieve efficiency in the fragment-docking calculations through utilization of dynamic programming. Finally, they plan to apply this procedure and test it on the MHC type I and II proteins. The research should lead to improved computerized schemes of docking for flexible ligands. This is a difficult problem that needs solutions in order to increase our understanding of protein-protein interactions, drug development and interactions among various biological molecules.

Scope: The proposal addresses an important problem in molecular modeling: docking of peptides to protein receptors and calculation of binding free energy. Both structural and energetic aspects of this problem are important. The ability to predict the structural aspects and the binding affinity of ligand-receptor interactions has important implications for molecular biophysics, as well as for the emerging area of structural bioinformatics. The principal investigator has published a number of papers on this and related topics, under current NSF funding. Prof. Sezerman has previously collaborated with Dr. Vajda. This project will promote a natural fit between the PI's group and the group of his collaborator at Sabanci University in Turkey. The areas of expertise and the resources in these two places are highly complementary. The project fits well with the objective of the Division of International Programs for encouraging mutually beneficial collaboration on research in important scientific problems.

DESCRIPTION (provided by applicant): Studying protein-protein interactions is crucial to gaining a better understanding of processes such as metabolic control, signal transduction, and gene regulation. Many biologically important interactions occur in weak, transient complexes that are not amenable to direct experimental analysis, even if both interacting proteins can be isolated and their structures determined. Thus it is important to develop computational docking methods that, starting from the structures of component proteins, can determine the structure of their complexes with accuracy close to that obtained by X-ray crystallography. Our multistage approach to docking starts with a rigid body search that samples billions of conformations on a grid using fast Fourier transform (FFT) correlation calculations, and includes post-processing to find near-native conformations among thousands of docked structures. Our methods were among the best in the latest CAPRI docking experiment, but results are still poor for antigen-antibody and some other types of complexes. The general goal of this proposal is the development of new algorithms that will improve the accuracy and reliability of multistage docking while retaining its computational efficiency, hence enabling the use of state-of- the-art methods for a web-based docking server. We will address three problems. First, Monte Carlo-based docking programs have demonstrated that searching for optimal side chain conformations in the interface can improve results. However, Monte Carlo is not a very efficient search, and such programs are computationally too demanding to globally explore the conformational space without a priori information on the structure of the complex. To remedy this we will develop efficient docking-specific side chain search algorithms within the framework of the multistage approach. This will include the identification of key side chains that are most important for recognition, thereby reducing the side chains' degrees of freedom. Second, accounting for electrostatics, desolvation, hydrogen bonding, and possibly experimental constraints improves docking results, but the use of such complex scoring functions reduces the numerical advantage provided by the FFT correlation approach. We will regain this advantage by developing a multi-property 5-dimensional FFT-based algorithm that can be used with scoring functions of arbitrary complexity without added computational costs. Third, docking results are particularly poor for antibody-antigen pairs. Our preliminary data indicate that results can be substantially improved by adopting asymmetric hydrophobicity potentials that account for the biophysics of interactions in particular classes of protein-protein complexes. In the first six months of the project we set up a new version of our ClusPro server, heavily used by experimentalists, which will include recent developments. The server will be continuously updated as our new algorithms become available. In addition to providing the server, we have already started development on a modular program library specific to the docking problem that will be freely accessible. PUBLIC HEALTH RELEVANCE: Studying protein-protein interactions is crucial to gaining a better understanding of processes such as metabolic control, signal transduction and gene regulation. Since many biologically important interactions occur in weak, transient complexes that are not amenable to direct experimental analysis, it is important to develop computational docking methods that, starting from the structures of component proteins, can determine the structure of their complexes with accuracy close to that obtained by X-ray crystallography. The general goal of this proposal is to further improve the accuracy of multistage docking while retaining its computational efficiency, thereby enabling the use of state-of-the-art methods in protein docking servers available to the scientific community.

Several research projects in this proposal focus on environmental pollutants that bind to specific protein receptors and impact regulatory pathways, resulting in developmental and reproductive abnormalities. The bioinformatics and molecular modeling research support core will facilitate the use of modern computational tools to study these interactions at the molecular level. Software and expert help is offered in four different areas: (1) bioinformatics, (2) molecular modeling and visualization, (3) protein structure prediction by homology modeling, and (4) analysis of receptor-ligand interactions. The bioinformatics tools include screening databases for similar sequences, aligning sequences for evolutionary inferences, and constructing evolutionary trees. The core will provide a variety of molecular modeling and visualization tools for the analysis of DNA and protein structures. In particular, software packages and modeling expertise will be provided for elucidating the tertiary structure of proteins that exhibit a reasonable degree of homology (at least 30% sequence identity) to a protein with known 3D structure, thereby providing potentially very valuable structural information in several research projects Protein structures, obtained either experimentally or by homology modeling, will be used to construct putative models of complexes with ligands of interest. Docking algorithms are provided for small organic molecules, flexible peptides, and proteins The analysis of predicted receptor-ligand interactions will suggest further experiments, and may help to elucidate the mechanism of action of the ligands studied. The core will have programs to search entire databases of organic molecules for ligands with high affinity for a given receptor.

Acquisition of Computing Infrastructure for Bioinformatics Research and Education

A grant has been awarded to Dr. Zhiping Weng at Boston University to establish a high-performance, cost-effective computing infrastructure to support the research and educational needs of the University's Bioinformatics Program. Dr. Weng and six co-investigators of this grant form the computational core of the program. Their research spans structural proteomics, functional genomics and gene expression/regulatory networks. On a daily basis they work with all known sequences, all available genomes, all known protein structures, transcriptional levels of all genes in an organism, and the like. Their computational needs far exceed currently available facilities. Boston University is among the first to establish a graduate degree program in Bioinformatics. The program is currently funded by an NSF IGERT grant and has been growing very rapidly. The large number of students in the program places great demand on computing resources.
This NSF grant will allow for the purchase of a computer cluster dedicated to Bioinformatics research and educational activities at Boston University. The cluster will have the following components: a file server with 1 terabyte storage, an 80-node compute server with dual Intel 1GHz processors, 2 GB RAM, and 106 GB storage per node, a backup file server and a web server. A fast giga-bit network will connect the file server and the compute server. Such a cluster coincides ideally with the high-throughput, large-scale nature of Bioinformatics research programs.
The computer cluster will be able to support innovative research in biomolecular systems including functional genomics, structural proteomics, and cell systems analysis. It will enable the faculty, postdoctoral researchers, graduate and undergraduate students to attack research questions that require months of CPU time, and which could not otherwise be addressed within the constraints of currently available facilities. It will also greatly strengthen our Bioinformatics Program at Boston University, which is dedicated to training a new generation of scientists and biotech industry leaders capable of implementing high-performance computational solutions to the biological problems of the 21st century.

DESCRIPTION (provided by applicant): Support is requested for a conference MODELING OF PROTEIN INTERACTIONS IN GENOMES. The conference will be held at the Lightsey Conference Center in Charleston, SC, on June 16-19, 2001, and will focus on the computational procedures for reconstruction and study of the network of connections between proteins in genomes. The networks of connections between proteins have to be built by a combination of experimental studies, knowledge-based methods, sequence analysis, and structural approaches. The goal of this conference is to discuss realistic aims, computational methods development, and applications. The following topics will be covered: protein-protein docking (structure prediction and binding simulation), energetics of protein-protein association (energy landscapes, stability of complexes), protein structure-function relationships (functional assignment of protein structures, binding site prediction), interacting and noninteracting proteins (structure-based and sequence-based approaches, pathways, experimental approaches), modulating protein-protein interactions by small ligands, and the design of small molecules that can play some roles of native proteins in specific interactions. The conference will be relatively small, with approximately 100 participants. The invited speakers will represent the protein docking community, including Michael Sternberg, ICRF, UK; Joel Janin, CNRS, France; and Ruth Nussinov, NIH, as well as researchers working on the energetics and kinetics of protein-protein association; e.g., Barry Honig, Columbia University; Andrew McCammon, UCSD; Harold Scheraga, Cornell University; and Peter Wolynes, UCSD. The relationship between protein structure and function will be addressed (Jeffrey Skolnick, Danforth Center; and Adam Godzik, Burnham Institute). An entire section will be devoted to the modulation of protein-protein interactions by small molecules (e.g., Arthur Olson, Scripps Institute; Brian Shoichet, Northwestern University, and researchers from the pharmaceutical indutry), and another to the experimental analysis of protein-protein interactions, including two-hybrid studies and the identification of pathways, including the current state of pathway libraries. A poster session will be held on one of the conference days.

The conference, to be held in June 2001 in Charleston, SC, will focus on computational procedures for developing and studying the network of connections between proteins. The networks of connections between proteins in genomes have to be built by a combination of experimental studies, knowledge-based methods, sequence analysis and structural approaches that are both fast and insensitive to inaccuracies. The Advisory Committee is committed to attending the conference, which will be on the order of 100 participants. The PI is committed to involving junior researchers and under-represented groups as speakers and participants in this conference.

A major challenge in structural genomics is the elucidation of enzyme active sites from molecular structure. The characterization of the ligand/substrate binding site provides the basis for a number of applications, including the design of inhibitors for the analysis of biochemical and signal transduction pathways, drug design, and protein engineering aimed at altered specificity or catalytic activity. The primary sources of information on specific molecular interactions are the structures of the enzyme (or its homologues) co-crystallized with various ligands (substrates, cofactors, inhibitors, products, and transition state analogs). Although such structures are available for over 70% of the enzymes currently in the Protein Data Bank, collecting all binding site information for a particular protein requires substantial efforts.

The goal of this proposal is to develop a computational resource and database, called Predicted and Consensus Interaction Sites in Enzymes (PRECISE), which will provide query and visualization tools for the comparative analyses of the interactions extracted from all relevant structures. For each enzyme, the web-based analysis tool will determine the consensus binding site, obtained by aligning all homologous sequences, identifying the residue positions that are important for the binding of any ligand, and assessing the roles of amino acids at these positions. A dynamic retrieval system will help to examine individual interactions, and to generate various statistics. In addition to the interaction extracted from enzyme structures, the database will also store putative interactions, based on mutation studies or theoretical predictions. Such interactions can be submitted to the website and checked for consistency with all the interactions extracted from structural data. The submissions will be stored in a separate table, adding a predicted interaction component to the database.

The database will also house the results of applying computational solvent mapping, a novel binding site prediction method, to a large set of well-characterized enzymes. The method moves molecular probes - small organic molecules or functional groups - on the protein surface in order to identify the positions that bind the highest number of different probes. It was shown that such consensus binding occurs at major subsites of the enzyme binding site, and the amino acid residues that interact with the probes also bind the specific ligands of the enzyme. Thus, computational solvent mapping can be used for the identification and characterization of enzyme binding sites. In addition to providing solvent mapping results for a set of well-characterized enzymes, an e-mail server will be set up, so that solvent mapping calculations can be requested for any enzyme. It is expected that the PRECISE website will become a central depository of enzyme binding site information, will have a major impact on enzymology research due to the solvent mapping web-server, as well as provide an important educational resource for studies in biochemistry and molecular biology.

DESCRIPTION (provided by applicant): Computational solvent mapping methods place molecular probes - small molecules or functional groups - on a protein surface in order to identify the most favorable binding positions. X-ray crystallography and NMR show that organic solvents with different sizes and polarities cluster at a limited number of sites on a protein. A recently developed mapping algorithm reliably identifies the consensus sites at which different probe molecules bind. These sites are in good agreement with the available experimental data. A very important result is that the consensus sites in enzymes are major subsites of the substrate binding site, and the amino acid residues that interact with the probes also bind the specific ligands (substrates, inhibitors, and transition state analogs) of the enzyme. Thus, computational mapping can be used for the identification and characterization of functional sites. The approach is less sensitive to variations in the structure of the protein than docking methods, and is remarkably robust against changes in the algorithm and energy parameters. The goals of this proposal include the development of software for high-throughput automatic mapping, validation of the approach by the mapping of a number of well understood enzymes, and application of the method to as many poorly characterized enzymes as possible. The results are expected to provide a substantial body of new information on enzyme binding sites, and at the very least should suggest which residues should be studied by site-directed mutation experiments. Solvent mapping will also be tested for its ability to identify functional sites in other types of proteins. A reliable mapping method will be particularly useful, as structural genomics approaches are likely to produce structures for an increasing number of poorly characterized proteins, and there are very few computational methods for identifying functional sites on the basis of protein structure, As a second application, amino acids will be used as probes, and the relationship between favorable binding positions of individual amino acid residues and their actual positions in bound peptides will be studied.

DESCRIPTION (provided by applicant): Support is requested for the multidisciplinary conference MODELING OF PROTEIN INTERACTIONS IN GENOMES 2005, to be held on 24-27 June, 2005 in Lawrence, Kansas. The meeting will involve a combination of biological, physical, and computational scientists, and will focus on computational procedures for reconstruction and study of the networks of connections between proteins in genomes, including experimental studies, knowledge-based methods, sequence analysis, and structural approaches. The specific aims of the Modeling of Protein Interactions in Genomes Conference 2005 are as follows: (1) discussion of the structural aspects of protein interactions, including docking methods; (2) discussion of current and emerging methods of determining protein binding sites using both sequence-based and structure-based approaches; (3) modeling of protein-protein interactions on the basis of a diverse set of experimental techniques (e.g., mass spectroscopy, cryo-electron microscopy, cross-linking, NMR, etc); and (4) solving computational problems associated with the analysis of protein interaction networks from large-scale screening experiments (primarily two-hybrid essays and mass spectrometry). Experts from various areas of protein interactions were selected to serve on our Advisory Committee and also to present at the conference. Further presentations will be invited as well as selected by the Advisory Committee from the submitted abstracts. A poster session will be held on one of the conference days. The conference will be relatively small - approximately 100 participants. The organizers will pay special attention to attracting young investigators as speakers and poster presenters. We will encourage participation of students and postdoctoral fellows. An important aspect of the conference organization will be our effort to diversify the participation by attracting women and minority scientists. This will primarily achieved by contacting faculty members working in the areas of Biophysics and Biochemistry in minority-serving institutions.

[unreadable] DESCRIPTION (provided by applicant): Support is requested for the multidisciplinary conference Modeling of Protein Interactions 2007 (MPI2007) to be held September 30 - October 2, 2007 in Lawrence, Kansas. The meeting will involve a combination of biological, physical, and computational scientists, and will focus on challenges and available methodologies in the determination of protein complexes, including experimental studies, knowledge-based methods, sequence analysis, and structural approaches. The primary goal of MPI2007 will be to critically explore the availability of tools for improving the reliability of protein interaction information, focusing on broadly defined template-based modeling of protein-protein interactions. Since the analysis is based on the structures of proteins and incorporating additional information attempts to contribute to the understanding of their cellular roles, a large scale reconstruction of protein interactions will build on the results of the Protein Structure Initiative. Some of the fundamental questions we will discuss are as follows. Will protein docking evolve in a way similar to structure prediction? In other words, will knowledge-based methods provide the most useful information, and will be there enough protein-protein templates? Does a large-scale concerted effort to determine protein-protein templates, similar to PSI, make sense, at least in principle? Is the field mature enough to meet such a challenge? How can prediction tools be combined with a diverse set of experimental techniques (e.g., mass spectroscopy, cryoelectron microscopy, cross-linking, NMR, etc) to improve reliability? [unreadable] [unreadable] Experts from various areas of protein interactions were selected to serve on the Advisory Committee and to present at the conference. Further presentations will be selected by the Advisory Committee to cover all major methodologies of protein complex determination and modeling. A growing number of highly qualified women are working in the areas covered by the conference and the female speakers will be represented in each section. A poster session will be held on one of the conference days. The conference will be relatively small - approximately 100 participants. The organizers will pay special attention to attracting young investigators as speakers and poster presenters. Participation of students and postdoctoral fellows will be encouraged. Efforts will be made to diversify the participation by attracting minority scientists and graduate students by contacting faculty members working in the areas of Biophysics and Biochemistry in minority-serving institutions. The concept of the cell as a collection of multisubunit protein machines that determine its behavior in normal and diseased states is emerging as a cornerstone of modern biology. The planned Modeling of Protein Interactions 2007 (MPI2007) meeting will involve a combination of biological, physical, and computational scientists, and will focus on challenges and available methodologies in the determination of protein complexes, including experimental studies, knowledge-based methods, sequence analysis, and structural approaches. [unreadable] [unreadable] [unreadable] [unreadable]

? DESCRIPTION (provided by applicant): The proposal requests the renewal of the grant Computational Mapping of Proteins for the Binding of Ligands. Mapping globally samples the surface of target proteins using fragment sized molecular probes. The general goals of mapping are determining binding hot spots, i.e., regions of proteins that are major contributors to the binding free energy, and identifying fragments with preferential binding to these hot spots. The main advantage of studying hot spots is that they are more conserved than binding sites are. We pursue four aims. First, we improve the efficiency of flexible mapping by performing side chain search directly within the global mapping algorithm, and extend the algorithm to models with flexible loops and to homology models. The method will be used for large scale mapping calculations to answer interesting biological questions such as the existence of druggable cryptic sites in the kinome. Second, we develop an effective combination of computational and experimental methods for the identification of fragments binding to a given hot spot, and working with collaborators attempt to find fragment hits for a number of drug targets, in the process generating fragment binding data needed for improving computational methods. Third, we develop an algorithm for virtual fragment screening in order to reduce the number of fragments that need to be experimentally tested. The resulting methods will provide direct input for fragment based ligand discovery (FBLD), thereby reducing the high costs of the approach and making it more accessible to academic laboratories. Fourth, we will further improve the accuracy of mapping by explicit modeling of solvation. The method is based on decomposing the protein into flexible side chains with multiple conformers, the rest of the protein, two copies of a probe in a number of conformations, and two water molecules. We sample the interaction energies for all feasible relative orientations of all pairs using very efficient fast Fourier tranform correlation methods, and store the detailed energy grids in lookup tables, compressed using wavelet transforms. Due to the additivity of pairwise interaction energies, the energy of any conformation can be quickly evaluated by adding pre-calculated internal energy components to interaction energies, all extracted from lookup tables. The use of energy tables will speed up the calculation of partition functions, refinement of fragment positions using Monte Carlo simulations, and determining escape times by stochastic roadmap simulation. All new algorithms will be implemented in our FTMap server (http://ftmap.bu.edu), which already has over 1200 registered users.

Protein-protein interactions are integral to virtually all biological pathways. Predicting these interactions and the function of the protein complex in key to understanding how biological pathways function. Detailed multistage docking algorithms, which starts from the unbound structures of two proteins, can determine the structure of the protein complex. The docking server, ClusPro, strives to make these docking algorithms accessible to researchers. However, the current refinement stage is computationally too demanding for use in an online server, and hence is replaced by simple energy minimization. The ClusPro team will develop methods to perform side chain search within a traditionally rigid body docking algorithm, and to calculate escape times from each energy funnel by stochastic roadmap simulations. These methods will provide more efficient refinement and will help to identify near-native models, thereby improving the reliability and accuracy of predictions. The server will be implemented on a number of platforms, including supercomputers and multi-core desktops.

ClusPro already has over 4500 users and runs over 1000 jobs per month. In 2011, 164 papers used models generated using the server to study problems in biology, biochemistry, and biotechnology. The upgraded server, with its simple user interface, will be particularly useful to experimentalist with no extensive computational experience, who will be able use it for generating models of protein interactions to explain their data. Graduate students will be trained to optimally combine high performance structure prediction algorithms with experimental data from a variety of low-resolution or non-structural techniques. Docking methods are also being incorporated into undergraduate courses to teach the biophysical principles of molecular recognition.

Proteins are the major building blocks of the cell. Many proteins perform their function by interacting with other proteins. In a typical cell hundreds of thousands of different protein interactions take place. Characterizing these interactions helps elucidate how living organisms function at the molecular level, contributes towards the development of treatments against diseases such as cancer and facilitates the design of novel bio-inspired materials. Detailed understanding of protein interaction mechanisms requires determining the three-dimensional structures of protein-protein complexes. These structures are very difficult to obtain using experimental techniques, thus, computational approaches can be very useful. The team of Kozakov, Paschalidis, Vajda and Vakili has developed algorithms and software that, according to the worldwide evaluation experiment CAPRI (Critical Assessment of Predicted Interactions), are among the best for predicting the structures of protein-protein complexes. These methods have been implemented in the fully automated docking server ClusPro, which is free for academic use, and has over 10,000 regular users. However, the current tools are computationally too demanding to serve such a large user base or to model protein interactions on a genomic scale. The goal of this project is to use rigorous geometrical and biophysical principles to substantially improve the efficiency of docking algorithms while retaining the accuracy of the generated models. Faster modeling of protein complexes will lead to better understanding of fundamental biological questions at both the cellular and system levels and will facilitate biochemical, biomedical, and biotechnology research. In addition, the methods will be used in training graduate students and teaching undergraduate and high school students.

The protein-docking problem is to computationally determine the 3-dimensional (3D) structure of the complex formed by two unbound proteins, given their individual 3D structures, by finding the global minimum of an energy-based scoring function. If one of the proteins, considered the receptor, is in fixed position and orientation, the search space includes the 6D rotational/translational space of the other protein, considered the ligand, as well as additional degrees of freedom that represent the flexibility of the two proteins. Since the energy function has a large number of local minima separated by high barriers, the minimization problem is extremely challenging. The proposed project aims to draw on the innovative manifold and optimization-based approaches that the group has developed for various components of docking protocols in order to (i) put the algorithms on a solid theoretical footing, (ii) study their performance and behavior more rigorously, and (iii) develop generalizable features of the algorithms that can be applied to other application domains. The work will focus on four key algorithms. First, the Fast Fourier Transform on Manifolds (MFFT) will be studied, which enables global systematic search in the space of rigid body motions of one protein with respect to the other. The computational complexity of MFFT will be analyzed and numerical performance tests for different bandwidth settings will be conducted. Second, underestimation-based sampling techniques will be explored, acting on the same manifold and targeting a medium-range search to refine MFFT solutions. A variety of underestimators will be developed and tested. Third, manifold-based local optimization approaches for flexible protein optimization will be developed. In each case, different manifold parameterizations and different minimization approaches will be compared. Finally, a new formulation of side-chain packing and other problems arising in protein docking will be developed using principles from the theory of Markov Random Fields. In this context, different solution approaches will be developed involving the Maximum Weight Independent Set (MWIS) problem and generalized belief propagation. The algorithms developed will be released as an open source software library to be used both for protein docking and in other application domains.

Each living cell is packed with proteins that continuously interact with each other in response to envirtonmental and other signals to control the cell's growth and eventual fate. The analysis of such interactions is crucial for understanding signal transduction pathways that drive cell differentian during development and throughout the life an an organism. These pathways are often involed in disease progression, such as cancer, so a broader impact of understanding these pathways could be the design of drugs to modulate such pathways. Moreover, such an understanding of protein protein interactions may aid the develpment of protein-based biomaterials. Current technologies exist for providing a blueprint for large protein interaction networks, but a deeper understanding requires detailed structures of the complexes formed by protein pairs or partners. Structural information is frequently difficult or even impossible to obtain by experimental tools, emphasizing the need for computational approaches. Vajda and Kozakov at Boston University have developed the web based server ClusPro for predicting the three dimensional structures of protein-protein complexes. According to the worldwide experiment CAPRI (Critical Assessment of Predicted Interactions), ClusPro consistently has been the best protein-protein docking server. It has over 5000 registered users and performs around 3500 docking calculations each month. Structures generated by the server have been reported in over 350 research papers. The goal of this project is to develop the next generation of ClusPro that will be able to optimally utilize the vast amount of data accumulated in public bioinformatics databases in order to improve the reliability and accuracy of the predicted structures. Since the server is used primarily by biological and chemical scientists who may not have expertise in the use of current bioinformatics tools, the new version will provide convenient access to state-of-art methods. Integration of bioinformatics and computational biophysics approaches, in combination with experimental validation, will result in a unique and powerful research tool. The increased availability of protein complex structures will have a major impact in many areas of biology, biochemistry, and biotechnology. All software developed will be released free of charge for academic and governmental use. In addition, the project will be used to train a new generation of graduate students, who will be able to optimally combine data from a variety of experimental and bioinformatics techniques with high performance computing. The use of the server will also be incorporated into undergraduate courses to teach aspects of bioinformatics and biophysical principles of molecular recognition.

The current version of ClusPro systematically samples the conformational space of a target protein complex, and scores the structures using physics based energy functions. The major shortcoming of this approach is that it performs docking without consideration of the large body of interaction, sequence, structural, and experimental information available in public databases. Rather than simply generating and scoring docked structures of the starting proteins, the new approach will use their sequences to collect all orthologs to identify the globular domains and inter-domain regions that may contain short linear motifs (SLIMs). Domains that occur in a number of interacting protein pairs are likely to mediate the interaction, enabling the identification of the specific domains that interact with domains or SLIMs of the partner protein. Superimposing the domains from different orthologs identifies the conserved regions. Since the interface in complexes is sequentially and structurally more conserved relative to the rest of the protein, for the case of domain-domain interactions docking just the key segments generally yields near-native docked structures. Docking key segments from different orthologs and selecting consensus models improves the reliability of predicted structures. This approach substantially improves the docking of proteins that have flexible loops, and will be extended to the docking of homology models. If the interface includes a SLIM, docking structures of peptide fragments that contain the motif extends the method to proteins with interactions mediated by flexible or unstructured regions. The algorithms to be developed will be implemented in ClusPro, which will provide substantial new information on protein-protein complexes. ClustPro software can be accessed at http://cluspro.bu.edu.

? DESCRIPTION (provided by applicant): The research in our lab focuses on molecular recognition using computational methods and follow-up validation experiments. Our primary target areas are (1) protein-protein docking and (2) exploring binding properties of proteins by computational solvent mapping. Protein docking methods are needed because many important interactions occur in weak, transient complexes that are not amenable to direct experimental analysis. We have developed ClusPro, the best docking server currently available. While the server is heavily used, with over 350 research papers reporting models constructed by ClusPro, the methodology has several limitations. First, global docking of relatively rigid proteins usually generates structures within 10Å interface RMSD from the native complex, but selecting and refining the best models frequently fail. Second, the methods are less accurate when docking peptides, proteins with flexible loops or unstructured regions, or homology models. Third, no reliable method is available for determining whether a docked structure represents a stable complex, and for calculating its binding free energy with any reasonable accuracy. Fourth, even these imperfect methods are too slow for proteome-wide analyses. We expect to address and solve all these problems. In addition, a new approach, based on pre-calculated pairwise interactions, will be developed for modeling complex systems, including aggregation and crowding effects. The second application considered in the proposal, computational solvent mapping, globally samples the surface of target proteins using fragment sized molecular probes. The general goals of mapping are determining binding hot spots, i.e., regions of proteins that are major contributors to the binding free energy, and identifying fragments with preferential binding to these hot spots. The main advantage of studying hot spots is that they are more conserved than binding sites are. We will improve the efficiency of flexible mapping by performing side chain search directly within the global mapping algorithm, and extend the algorithm to models with flexible loops and to homology models. The method will also be used for large scale mapping calculations. We will develop an effective combination of computational and experimental methods for the identification of fragments binding to a given hot spot, and working with collaborators attempt to find fragment hits for a number of important drug target proteins. The ultimate goals of this research are developing algorithms for virtual fragment screening in order to reduce the number of fragments that need to be experimentally tested, and expanding the method to provide direct input for fragment based ligand discovery (FBLD), thereby reducing the high costs of the approach and making it more accessible to academic laboratories and small companies.